The paper deals with the problem of identifiability of mean-value models of turbocharged IC engines. A way of reducing such models to linear regressions using differential algebra is presented. The conditions of the global identifiability and the persistent excitation are formulated in explicit form for a given set of sensors. It is accompanied with the discussion how to reduce the set of sensors required for engine identification. As a side product it is shown how to use regressors derived for estimation and/or verification of some static nonlinearities using dynamic test data. The results obtained can be used for both model-based FDI (fault detection and isolation) and engine control purposes but mainly for automatic generation of engine test plans and identification.